[Lf] Re: lf digest, Vol 1 #147 - 3 msgs

[Vittorio De Tomasi]

Hi to all,
I follow up the discussion about how to digitize and compress audio data
on a narrow bandwidth.

Frank is right when he says that one could sample at high frequency, and
keep only one sample out of four (just to make an example). In this case
you have no aliasing, because you are operating on a narrow band signal.
There is however another way to do this. 

Let us say, as you described in your example, that you want only 200 Hz
of bandwidth, say [800-1000] Hz. Instead of placing a bfo and a mixer
between the filter and the ADC, simply use an ADC sampling at 400 Hz.
You must check that the sample-and-hold before ADC can pass the desired
bandwidth (I bet it will at such low frequency: this is probably not
true on HF!), and that no anti-aliasing filters exist after the narrow
band filter: in other words, leave the ADC free to generate plenty of
aliasing!

Now what happens is that the ADC samples correctly the frequency
interval [0-200], wrap arounds [200-400] into [200-0] interval, aliases
without wrap-around [400-600], then again wraps around [600-800] into
[200-0], and eventually correctly aliases [800-1000] into [0-200]. Since
the signal is bandpass, the aliased subbands give no contribution to the
ADC conversion...

The problem is how to implement this on a PC. I do not recommend using a
modified sound blaster, because you cannot easily assure that this could
be done on other cards (if you really want to experiment, and want to
modify a sound card, either look for cards using AD1845 or similar,
whose data sheets are easily available on Analog Devices web site, or
use an external ADC connected to SPDIF input if available). Be careful
also when specifying the sound card sampling rate. Infact
Microsoft-Intel specifications about multimedia PC prescribe that
*official* sampling frequencies for a so called "Multimedia level 2"
sound card are 11,025 22,050 and 44,100 Hz with 16 bit of resolution.
Other frequencies could be available on some cards, but you never know
what happens with another card...

If you want to use the sound card, sample at 11,025 Hz and put some DSP
code in the PC. You have to band-pass the data with a FIR filter (40-60
taps will do a brick-wall filter), then resample using an appropriate
method. As an alternative use the fast wavelet transform (AKA subband
filtering) on input data, so you can easily extract the subband
[689-1378] Hz readily undersampled by the fast wavelet transform.

An alternative is building a dedicated ADC connected to the PC using
serial (not recommended, UARTs have small buffers), parallel (used as
EPP port, so you can take advantage of DMA transfer mode!), or USB
ports. This solution has the advantage that it allows placing ADC close
to the audio source, and leaving the PC far from the radio.

I know that Cypress and Microchips recently released some
microcontrollers with USB support, they could be a good platform to
develop a dedicated ADC. As an alternative, I have somewhere a reference
to an ADC that can be directly interconnected to USB bus: I'll look for
the URL tomorrow.

Abut data compression, one could consider A-law encoding (or maybe u-law
encoding for people living in Europe :-) ) applied to each sample.
Basically, you digitize data with 12 bits, then amplitude is compressed
using an appropriate function into 8 bit. Dedicated ADCs exist (there is
one in each ISDN telephone!), but A-law compression is easily done also
with DSP methods. The small distorsion introduced by compression is
probably negligible for our purposes.

I suggest however not to waste time with audio compression methods,
other than dynamic compression: I think that we cannot do better than
the industries in this field (and I am not thinking about media
industry, with MP3, MPEG2, and so on... oil industry for example spent a
lot of money and efforts to develop sophisticated methods for data
compression of noisy seismic data, with less than marginal results).

vy 73

Vittorio IK2CZL


> Today's Topics:
> 
>   1. Re: Digitized Audio (Frank Gentges)
>   2. Re: lf digest, Vol 1 #146 - 1 msg (Vittorio De Tomasi)
>   3. Re: Re: lf digest, Vol 1 #146 - 1 msg (Frank Gentges)
> 
>   ------------------------------------------------------------------------
> 

< following text compressed to zero bytes lenght >


-- 
*************************************************************************
Vittorio De Tomasi         ik2czl at amsat.org
Home page:                 http://space.tin.it/scienza/vdetomas 
My DSP page:               http://www.freeyellow.com/members/padan

"Wir muessen wissen; wir werden wissen" (David Hilbert)